How Many Spoons Does Your AI Coding Tool Cost?
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
AI coding agents can reduce developer effort, but vendor lock-in creates a hidden energy tax: teams get stuck on one vendor's model, can't benefit from faster-moving open-weight alternatives, and lose time debugging opaque failures. The open-weight model market is accelerating rapidly — Kimi K3 (2.8T parameters), DeepSeek V4 (MIT license), and Qwen 3.8-Max all shipped within days of each other, with open-weight models now 4–7 months behind closed frontier labs (down from 6–10 months) and accounting for ~45% of tokens on major routing platforms. Open-source agents that support bring-your-own-API-keys and multi-model routing let teams adopt new models without rebuilding workflows. Kilo Code is presented as an example: Apache-2.0/MIT licensed, supporting 500+ models with zero inference markup, available in VS Code, JetBrains, CLI, and cloud, with 3M+ developers and 40 trillion tokens processed.
Table of contents
The tax nobody puts a name onThe ground is moving faster than lock-in can trackWhat changes with an open agentBuild for the tax that’s actually thereQuestions this post answers
How far behind are open-weight models compared to closed frontier AI labs on benchmarks?
Open-weight models are currently measured at four to seven months behind closed frontier labs on hard benchmarks, down from six to ten months for most of the prior year, while running at a fraction of the cost. Chinese open-weight providers alone account for roughly 45% of tokens flowing through major routing platforms, up from under 2% a year earlier. Teams choosing between open-weight and closed models for AI coding workflows track this gap on daily.dev.
What are the largest open-weight model releases available for AI coding use?
Moonshot AI open-sourced the full weights of Kimi K3, a 2.8-trillion-parameter model described as the largest open-weight release to date. DeepSeek V4 reached general availability under an MIT license the same week, and Qwen 3.8-Max shipped days after that — all within a short window, illustrating how quickly the open-weight market is moving. Developers evaluating open-weight models for coding agents keep up with releases like these on daily.dev.